Combustion and Flame 144 (2006) 710–729

www.elsevier.com/locate/combustflame

Combustion kinetics of coal chars

in oxygen-enriched environments
Jeffrey J. Murphy 1 , Christopher R. Shaddix ∗
Combustion Research Facility, Sandia National Laboratories, Livermore, CA 94550, USA
Received 15 September 2004; received in revised form 5 July 2005; accepted 22 August 2005
Available online 16 November 2005

Abstract
Oxygen-enhanced and oxygen-fired pulverized coal combustion is actively being investigated to achieve emis-
sion reductions and reductions in flue gas cleanup costs, as well as for coal-bed methane and enhanced oil recovery
applications. To fully understand the results of pilot scale tests and to accurately predict scale-up performance
through CFD modeling, accurate rate expressions are needed to describe coal char combustion under these uncon-
ventional combustion conditions. In the work reported here, the combustion rates of two pulverized coal chars have
been measured in both conventional and oxygen-enriched atmospheres. A combustion-driven entrained flow reac-
tor equipped with an optical particle-sizing pyrometry diagnostic and a rapid-quench sampling probe has been used
for this investigation. Highvale subbituminous coal and a high-volatile eastern United States bituminous coal have
been investigated, over oxygen concentrations ranging from 6 to 36 mol% and gas temperatures of 1320–1800 K.
The results from these experiments demonstrate that pulverized coal char particles burn under increasing kinetic
control in elevated oxygen environments, despite their higher burning rates in these environments. Empirical fits
to the data have been successfully performed over the entire range of oxygen concentrations using a single-film
oxidation model. Both a simple nth-order Arrhenius expression and an nth-order Langmuir–Hinshelwood kinetic
equation provide good fits to the data. Local fits of the nth-order Arrhenius expression to the oxygen-enriched and
oxygen-depleted data produce lower residuals in comparison to fits of the entire dataset. These fits demonstrate
that the apparent reaction order varies from 0.1 under near-diffusion-limit oxygen-depleted conditions to 0.5 un-
der oxygen-enriched conditions. Burnout predictions show good agreement with measurements. Predicted char
particle temperatures tend to be low for combustion in oxygen-depleted environments.
© 2005 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

Keywords: Coal, char; Combustion; Kinetics; Oxy–fuel; O2 enriched

1. Introduction

Alternatives to the conventional air-blown, atmo-

spheric-pressure pulverized-coal (pc) combustion fur-
* Corresponding author. Fax: +1 925 294 2276. nace are being investigated to enhance energy ef-
E-mail address: [email protected] (C.R. Shaddix). ficiency, reduce greenhouse-gas and pollutant emis-
1 Current address: The Aerospace Corporation, P.O. Box sions, and minimize the size and capital cost of future
92957, Los Angeles, CA 90009, USA. coal-based powerplants. Coal gasification is receiv-
0010-2180/$ – see front matter © 2005 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustflame.2005.08.039
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 711

ing significant attention, both when integrated into a 1 atm [5–9]. Similarly, simultaneous char particle size
combined-cycle gas turbine powerplant (to form an and temperature statistics [10–13] (and, in two cases,
integrated gasification combined-cycle, or IGCC, sys- burnout times [11,13]) have been measured at various
tem) and when coupled to a synfuel chemical process- oxygen concentrations under pressurized conditions
ing plant to produce liquid automotive fuel or hy- for pulverized coal particles. In addition, Tidona mea-
drogen. Pressurized combustion of coal is also being sured char particle temperatures and burnout times
explored, especially in the form of pressurized flu- for laser-ignited pulverized coal particles surrounded
idized bed combustor (PFBC) technology. However, by room temperature, high-O2 fraction gas at 1 atm
one of the most promising near-term alternatives to and under pressurized conditions [14]. Unfortunately,
conventional pc combustion is atmospheric-pressure for all the studies using variable reactor pressures
pc combustion in mixtures of oxygen and recircu- but one [10], the investigators did not use a com-
lated flue gas. This technology could be employed in mon char source for the separate experiments, causing
new, smaller, stand-alone powerplants, but it is espe- devolatilization-induced variations in the char proper-
cially attractive as a potential retrofit technology for ties to become convoluted with the fundamental oxi-
existing pc units. The CO2 -rich product gas from this dation process [15,16]. Furthermore, none of the pre-
process can be beneficially used for enhanced oil re- vious investigations at elevated oxygen partial pres-
covery or coal-bed methane applications [1] or can sures derived coal char kinetic rate parameters from
be sequestered in geologic reservoirs. This combus- the measurements.
tion method utilizes enhanced oxygen levels (typi- In the work reported here, an entrained flow reac-
cally 28–30 mol%) to maintain a suitable furnace exit tor with an optical particle-sizing pyrometer is used
gas temperature (at high temperatures the molar spe- to measure the joint temperature-size statistics of
cific heat of CO2 is approximately 67% larger than size-classified pulverized coal char particles burning
that of N2 ) and sufficient heat-transfer rates to steam at 1 atm both in conventional, reduced oxygen fur-
tubes. nace conditions and in oxygen-enriched furnace con-
For all of the above-noted technologies, a portion ditions. From these data, we make the first quanti-
of the coal char conversion can be expected to oc- tative determination of high-temperature pulverized
cur in a high-temperature environment with the partial coal char combustion rates in oxygen-enriched en-
pressure of oxygen exceeding 0.21 atm. This is even vironments. In addition, char sampling and analysis
true for coal gasification, whether air blown or oxy- also yield the first reported information on elemen-
gen blown, because significantly elevated pressures tal release rates as a function of burnout in oxygen-
(typically 20–60 atm) are used in these processes. As enriched char combustion.
a consequence of ongoing research and development Deriving meaningful burning rate parameters from
of these coal conversion technologies, there is inter- the measured temperature and size data is not trivial,
est in understanding and measuring the combustion especially over the wide range of gas compositions
properties of coal under pressure and under oxygen- and particle temperatures considered in this investi-
enriched (>21 mol% O2 ) conditions. Of course, for gation. Therefore, special care has been taken here
all of these processes some fraction of the char con- to implement the best-available correlations for the
version occurs at intermediate to low partial pressures high-temperature gas transport properties of gas mix-
of oxygen, so the existing, extensive knowledge base tures. In addition, linear, nonlinear, and conventional
of char combustion in reduced oxygen (<21 mol% ad hoc methods have all been fully explored in the
O2 ) environments at atmospheric pressure is also rel- current work for optimizing the fits of kinetic ex-
evant. Furthermore, two recent investigations of low- pressions to the data. Comparisons are made between
to intermediate-temperature char oxidation kinetics particle burnout calculations using the recommended
at elevated pressure [2–4] (using chars produced at kinetic parameters and measured burnout. However,
1 atm) have demonstrated that the intrinsic char oxi- in contrast to previous work, the kinetic parameters
dation kinetics are insensitive to the total pressure of are not adjusted to optimize the predicted burnout,
the system, depending only on the partial pressure of because the large uncertainty in making burnout pre-
oxygen. This result implies that char oxidation kinetic dictions does not justify an alteration of the kinetics
studies performed at a total pressure of 1 atm may be derived from the optical measurements.
applied to the understanding of oxidation at elevated
pressures, once proper accounting is made for differ-
ences in diffusion rates, boundary layer reactions, and 2. Experimental methods
char structure in the pressurized application.
Several previous studies have measured pulver- The coal combustion experiments were performed
ized char particle temperatures and burnout times in Sandia’s optical entrained flow reactor facility,
in high-temperature oxygen-enriched environments at whose general design and operation have been previ-
712 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729

Fig. 1. Schematic diagram of Sandia’s char kinetic entrained flow reactor, together with its particle-sizing pyrometer and char
collection probe.

ously described in the literature [17,18]. A schematic mately the same for the widely varying combustion
of the facility is shown in Fig. 1. The furnace op- conditions through the use of calibrated neutral den-
erates at 1 atm and uses a diffusion-flamelet-based sity filters.
Hencken burner to provide high-temperature gas that The precision in these single-particle measure-
rapidly heats the injected coal particles and acts as ments was estimated using linearized versions of the
the char oxidation medium. The use of pure gases to equations used in the analysis, coupled with noise
supply the Hencken burner allows a wide variety of estimates from the recorded signal traces. Precision
product gas mixtures to be produced over a range of in the temperature measurements is better than ±1%
temperatures. In particular, oxygen concentrations ap- (typically ±20 K), whereas the precision in the mea-
proaching 50 mol% may be produced, using any one sured velocity is ±2%. The particle size measurement
of a variety of diluents. has significantly lower precision, typically ±20%.
A particle-sizing pyrometer was used to simul- Calibration of the pyrometer was performed over
taneously measure the velocity, diameter, and tem- multiple temperature points using a high-temperature
perature of individual burning char particles at se- blackbody source (up to 1700 ◦ C), focused onto a
lected heights (i.e., residence times) within the reac- chopper wheel located along the furnace axis. Op-
tor. A helium-quench, water-cooled sampling probe tical pinholes attached to the chopper wheel moved
was used to collect char samples for chemical and through the focal plane of the light collection optics,
physical characterization. The collection optics of simulating emitting particles. This setup allowed a
the pyrometer project a particle image onto a coded small correction (typically 5%) to be applied to the
aperture. Laser light scattered from the particles is measured particle sizes, to account for the imperfect
used to trigger the data acquisition, ensuring that focus of the detection optics. A LabView data acqui-
data are taken only if the particle image is in focus. sition system was programmed to reject particle emis-
Emission signals are detected through 40-nm FWHM sion traces that indicate irregular particle trajectories
bandpass filters with center wavelengths of 550 and or overlapping particle signals.
700 nm. The use of these filters avoids contamina- Furnace oxygen concentrations of 6, 12, 24, and
tion of the pyrometry signals by line emission from 36 mol% were investigated with nitrogen as the dilu-
sodium (589.0 and 589.6 nm) and potassium (766.5 ent gas. Mixtures of hydrogen and ethylene were used
and 769.9 nm). The size and velocity of the particle is as the burner fuel, resulting in a water vapor con-
calculated from the recorded 700-nm emission signal centration of 14 mol% and a carbon dioxide concen-
using the geometry of the coded aperture [17]. Tem- tration of 4 mol% in the furnace gases. Flow rates
perature is calculated using a Wien’s law ratio of both of fuel and oxidizer gases to the burner were cho-
emission signals. The mean light (i.e., photon) inten- sen to maintain a common adiabatic flame tempera-
sity on the photomultiplier tubes was kept approxi- ture as the oxygen content in the burner exhaust was
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 713

trigger threshold level to avoid self-triggering intro-

duces a bias toward large particles. To overcome this
difficulty, a higher-power 17 mW linearly polarized
HeNe laser was incorporated to provide a stronger
laser trigger signal, and a polarization filter was used
in addition to the laser-line filter (1 nm FWHM) in
front of the trigger detector to discriminate against
natural particle emission.
Two characteristic North American coal sources
were used for these experiments, in collaboration
with investigations occurring at the CANMET Energy
Fig. 2. Measured gas temperature profiles along the center- Technology Centre on oxygen-enriched and O2 /CO2
line of the combustion-driven furnace for the investigated recycle coal combustion [22–24]. These coal sources
char combustion conditions. Optical and physical sampling are Highvale, a western subbituminous B coal that is
of char occurred over heights of 5–30 cm, with the maxi- extensively mined in Canada, and a commercial high-
mum sampling height depending on the oxygen content of volatile eastern United States bituminous coal blend
the flow. provided by Ontario Hydro, an electrical utility com-
pany. These coals were ground and sieved into a 106
varied, thereby providing a common gas temperature to 125-µm size fraction for use in Sandia’s entrained
profile along the furnace centerline. Three different flow reactor facility. The analyzed coal and ash prop-
burner temperatures were investigated, in addition to erties of the ground and sieved coal are given in Ta-
the variation in oxygen concentrations, as shown in ble 1. The subbituminous coal has a higher content of
Fig. 2. These burner conditions correspond to adia- volatiles, moisture, oxygen, calcium, and sodium, as
batic flame temperatures of 1600, 1800, and 2000 K. would be expected. The subbituminous coal is a low-
Gas temperatures along the centerline of the fur- sulfur coal, whereas the eastern bituminous coal is a
nace were measured with a fine-wire (76 µm) type R medium-sulfur coal. The proximate, ultimate, and ash
beaded thermocouple and corrected for radiative loss analyses of the eastern bituminous coal are typical of
[19]. The radiation corrections were calculated using a Pittsburgh seam coal [25], so it is presumed that this
the Collis and Williams correlation [20] for Nusselt is the original source of the coal.
number for cylindrical bodies and the best informa-
tion available on wire emissivity and the transport
properties of high-temperature gas mixtures [21]. Es- 3. Numerical analysis
timated uncertainty in the gas temperature measure-
ment is ±50 K. Unheated nitrogen carrier gas is used 3.1. Char burning rate
to convey the coal particles to the reactor, resulting in
centerline temperatures at the base of the furnace that The instantaneous burning rate of an individual
are well below the adiabatic flame temperature of the particle is determined from temperature, velocity, and
burner gases. Once the carrier gas has diffused into size information by solving the energy balance for the
the surrounding burner gases (at a height of approx- particle, assuming a spherical, homogeneous, reacting
imately 5 cm, which corresponds to the approximate particle surrounded by a chemically frozen boundary
height at which coal devolatilization ends), the center- layer (i.e., single-film model). Heat losses from con-
line gas temperature steadily decreases with height in duction and radiation are considered, as well as the
the furnace, on account of losses through the quartz effects of thermal inertia and Stefan flow [26]:
walls. For each set of reactor conditions, there were
4 to 10 sampling positions, spaced 1.27 or 2.54 cm dp ρp cvp dTp  
= −εσ Tp4 − Tw4
apart, at which optical data were taken. Typically, 200 6 dz
 
particles were measured at each position. Char sam- 2λ κ/2
ples were collected and analyzed for a selected subset − (Tp − Tg )
dp eκ/2 − 1
of these optical sampling positions. + qh. (1)
When first operating the flow reactor under oxy-
gen-enriched combustion conditions, the natural par- Here, q is the overall burning rate per unit ex-
ticle emission was found to be strong enough to trig- ternal surface area of the particle. The quantity κ =
ger the data acquisition system in the absence of the (−qdp /λ)i νi cg,i is a modified version of the Peclet
laser-scattering trigger signal. These self-triggered number (i.e., the ratio of the convective velocity of the
particles may be out of focus (leading to overesti- net mass leaving the particle surface to the diffusive
mates of the particle size), and simply raising the velocity of heat leaving the surface). This quantity
714 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729

Table 1
Chemical analysis of coal samples
Proximate Highvale Eastern bituminous
(wt%, as rec’d) (wt% dry) (wt%, as rec’d) (wt% dry)
Moisture 7.61 0.75
Ash 11.39 12.33 8.82 8.89
Volatile 37.15 40.18 34.91 35.17
Fixed C 43.89 47.47 55.52 55.94
Ultimate (wt% dry) (wt% DAFa ) (wt% dry) (wt% DAF)
C 60.70 69.23 77.33 84.87
H 4.01 4.57 5.08 5.57
O (by diff) 21.86 24.93 6.29 6.92
N 0.84 0.95 1.45 1.59
S 0.28 0.32 0.96 1.05
Higher heating value (as rec’d) (DAF) (as rec’d) (DAF)
MJ/kg 21.59 26.65 32.13 35.53
Btu/lb 9283 11,460 13,815 15,277
Ash analysis (wt% of ash) (wt% of ash)
SiO2 48.29 54.11
Al2 O3 21.31 32.74
Fe2 O3 4.14 5.60
TiO2 0.74 1.37
P2 O5 <0.10 0.08
CaO 12.01 1.23
MgO 1.09 0.73
SO3 4.91 0.92
Na2 O 3.27 0.40
K2 O 0.25 2.31
MnO 0.07 <0.01
a Dry, ash-free.

characterizes the correction to the heat-transfer equa- increases beyond 100 µm and as gas and/or parti-
tion due to Stefan flow (see Ref. [26]). The νi ’s are cle temperatures increase (e.g., for combustion in
analogous to stoichiometric coefficients νO2 = OF, oxygen-enriched atmospheres). On the other hand,
νCO = −2(1 − OF), and νCO2 = (1 − 2OF), where an analysis by Makino [28] suggests that the particle
OF is the oxidizer-to-fuel ratio, which is calculated temperature and the bulk oxygen content have little
from the CO2 /CO production ratio at the char surface. influence on the critical diameter for CO flame igni-
The conduction term in Eq. (1) is written assuming tion in the boundary layer.
a Nusselt number equal to 2, as is appropriate for a The CO2 /CO production ratio at the char particle
spherical particle surrounded by static flow. In fact, surface is determined from the correlation CO2 /CO =
η
the particles in the flow reactor experience some ve- A0 pO0 ,s exp(−B0 /Tp ) with the coefficients sug-
2
locity slip relative to the laminar gas flow, but this slip gested by Tognotti et al. [29]: A0 = 0.02, B0 =
is estimated to be approximately 0.1 m/s, yielding a 3070 K, and η0 = 0.21. Then, OF = (1 + ψ)/2,
particle Reynolds number of less than 10−6 . At this where ψ = CO2 /CO/(1 + CO2 /CO) is the fraction
low Reynolds number, the assumption of both heat of carbon that becomes CO2 . One could use semi-
transfer and mass transfer occurring in a nonconvec- global intrinsic kinetics, such as the three-step model
tive flow environment appears justified. proposed by Hurt and Calo [30], to calculate the
It should be noted that there is concern in apply- CO2 /CO production ratio. However, these intrinsic
ing the single-film char combustion model for the kinetic approaches fit the rate parameters to reproduce
combustion of particles larger than 100 µm in oxygen- correlations in the literature such as Arthur’s [31] or
enriched atmospheres. Steady-state char boundary Tognotti’s [29], so there is no benefit to applying the
layer modeling by Mitchell et al. [27] suggested intrinsic kinetic approach for this purpose.
that significant CO conversion to CO2 occurs in the The heat of reaction, h, is determined from the
boundary layer, with resultant impacts on particle value of OF and by assuming that the char has the
temperature and burning rate, as the particle size heating value of graphite. This assumption, previ-
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 715

where γ = −(1 − OF) and C = P /RTf is the aver-

age molar concentration of the gases in the bound-
ary layer. Given a trial value of q, Eq. (2) must be
solved iteratively because γ depends on the CO2 /CO
ratio, which in turn depends on pO2 ,s . This solution is
accomplished using a Newton–Raphson root-finding
method, with pO2 ,s constrained to a range of zero to
pO2 ,0 . If the root-finding method strays from these
bounds, or if convergence is not fast enough, a step is
taken using bisection.
Given experimental measurements of particle size,
Fig. 3. Measured heating value of the eastern bituminous
temperature, and velocity, the burning rate is deter-
coal char as a function of the extent of coal mass burnout
mined by iteratively solving Eq. (1) for q using the
in the entrained flow reactor. The heating value of graphite
is shown as a line for comparison. The graphite heating modified bisection algorithm described by Brent [34].
value is based on the JANNAF heat of formation for CO2 The solution method described in the preceding para-
(−394 MJ/kmol) at standard conditions. Both the fractional graph is used throughout the iteration to provide val-
coal mass release and the heating value are corrected to a ues of pO2 ,s , and thus CO2 /CO and h. This “nest-
dry, ash-free (DAF) basis. ed” approach to solving the equations is more compli-
cated than is required for calculating individual parti-
ously untested to the authors’ knowledge, was evalu- cle burning rates, but the additional complexity allows
ated during this study by measuring the heating value the same computer subroutines to be used with the
of the eastern bituminous char that was collected over nonlinear fitting techniques described later in this pa-
a wide range of burnout in the 12% O2 , intermediate per.
temperature furnace environment. The results, shown
in Fig. 3, indicate that the heating value of the de- 3.2. Char kinetic rate expressions
volatilized char (DAF mass release 0.4) is indeed
the same as that of graphite. Surprisingly, the mea- Historically, most kinetic data on burning coal
sured heating value of the char drops precipitously char particles have been interpreted using an nth-
above 80% coal conversion. These high-conversion order Arrhenius model of char combustion [25,35–
chars contain over 40 wt% ash, so it is presumed that 37]. In this representation, the global reaction rate
the low apparent heating values for these chars re- follows the expression
sult from either incomplete sample oxidation or some n ,
other error source in the calorimeter measurement. q = ks (Tp )pO 2 ,s
(3)
The commercial lab that performed the analysis did where n is the reaction order, Tp is the particle tem-
not see any unusual appearance in the final ashes and perature, and ks is a temperature-dependent rate coef-
had insufficient sample to analyze for carbon in ash ficient, assumed to follow an Arrhenius form:
after the calorimetry measurement.
Gas properties were calculated using the expres- ks (Tp ) = A exp(−E/RTp ). (4)
sions and molecular constants recommended by Ma-
son and co-workers [32]. The recommendations of Another kinetic expression that has been periodi-
Paul and Warnatz [33] were used to calculate thermal cally used to interpret char combustion kinetics is the
conductivity. The transient heating rate of the particle, Langmuir–Hinshelwood expression, which describes
dTp /dz, was approximated using the average particle competing adsorption and desorption reactions on
temperature data collected in the experiments. The the char surface [38,39]. The most commonly ap-
char particle density was assumed to be 550 kg/m3 plied form of this relation is the two-step Langmuir–
for the eastern bituminous coal and 700 kg/m3 for the Hinshelwood equation
Highvale coal, consistent with mean measured values. k2 k1 pO2 ,s
A specific heat of 2000 J/kg K was used for the chars q= (5)
k1 pO2 ,s + k2
from both coals.
Because the CO2 /CO ratio depends on the partial with the k’s, as before, expressing Arrhenius depen-
pressure of oxygen at the char surface (pO2 ,s ), the dence on the particle temperature.
gas-phase diffusion equation must be solved, As emphasized by Hurt and Calo [30], there are
    serious difficulties with both the nth-order Arrhenius
pO2 ,s pO2 ,∞ qdp
=γ + − γ exp − , (i.e., power law) and Langmuir–Hinshelwood rela-
p p 2CDO2 ,mix tions for describing carbon oxidation over a range
(2) of temperatures and partial pressures of oxygen.
716 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729

The Langmuir–Hinshelwood approach is theoreti- catalysts [42] (as usually assumed in the char com-
cally more attractive than the nth-order Arrhenius bustion literature) unless the external surface reaction
approach because it provides a description of the rate is negligible in comparison to the internal pore
O2 adsorption, CO (and CO2 ) desorption process reaction rate. This point has been previously noted by
that is understood to be the general mechanism by Mitchell [43] and Fortsch et al. [44].
which carbon is oxidized. However, the details of In principle, all three kinetic rate constants of the
the adsorption/desorption chemistry are poorly un- nth-order Arrhenius expression (A, E, and n) can be
derstood, particularly for heterogeneous materials determined by using linear regression to fit the burn-
such as coal char. Furthermore, the two-step, single- ing rate data directly to Eqs. (3) and (4) after applying
site Langmuir–Hinshelwood kinetic expression that a logarithmic transformation:
is most commonly used to describe char combustion
Ea
predicts trends in pressure dependence as a function ln q = ln A − + n ln pO2 ,s . (6)
of temperature that run counter to a large number RT
of measurements [30]. In fact, studies using thermo- In practice, this technique often results in a reaction
gravimetric analyzers (TGAs) to measure the low- order, n, close to zero, especially if the particles are
and intermediate-temperature oxidation of chars un- burning near the diffusion limit. Because the classi-
der both atmospheric pressure and pressurized condi- cal Thiele analysis constrains the apparent, Zone II
tions have found the nth-order Arrhenius expression reaction order to lie between 0.5 and 1 [36,42], the
to work well over a wide range of oxygen partial traditional practice has been to assume a value of n
pressures [4,30,40]. On first appearance, this result (usually n = 0.5) and then calculate ks coefficients
is inconsistent with Langmuir–Hinshelwood kinetics. from the data. Then, A and E are determined by fit-
However, Hurt and Haynes [41] have shown that in- ting a logarithmically linearized equation (4). With
clusion of heterogeneous reactivity distributions of this approach, the use of a fractional oxygen expo-
reasonable width into Langmuir–Hinshelwood kinet- nent requires the elimination of any burning rates at
ics results in apparent power-law kinetic behavior or above the calculated diffusion limit.
over a wide range of Zone I oxidation conditions. Hurt and Mitchell [37] pioneered the derivation
In addition, experimental trends in intrinsic oxidation of accurate kinetic constants from single-particle op-
rates at low temperatures are well approximated when tical data. When composing Arrhenius plots of ks
using the Hurt and Calo [30] three-step kinetic mech- versus 1/Tp , they noted that the single-particle rate
anism, including the reaction of molecular oxygen coefficients cluster into groupings corresponding to
with the oxygen–carbon surface complex to produce “characteristic curves” of the equations governing
CO2 and an additional surface complex. At high tem- gas-particle transport. The natural particle-to-particle
peratures, no existing kinetic approach can properly variations in reactivity, as well as experimental noise,
account for the low apparent reaction orders that have tend to scatter the data along these curves in a man-
been consistently measured, albeit with lower qual- ner that biases the regression and the resulting kinetic
ity data that spans a narrower range of variation of constants, particularly when data are collected at only
surface oxygen concentration than exists at low to in- a few different conditions (three sample heights and
termediate temperatures. two oxygen contents were investigated in their stud-
ies). To alleviate these difficulties, Hurt and Mitchell
3.3. Regression methodologies first removed all near-diffusion-limit datapoints from
consideration and then defined a set of characteristic,
In most previous studies of high-temperature char synthetic data points for each measurement condition
combustion kinetics, a given char kinetic expression and calculated the burning rates for these synthetic
was assumed to be applicable and a linear least- datapoints. Finally, linear kinetic fits were performed
squares regression was performed on the data, or (assuming a reaction order of 0.5) on these statisti-
some subset thereof, to optimize the relevant kinetic cally representative burning rates.
constants. This procedure is complicated by Zone II Linear fitting techniques have a major difficulty
particle combustion, where both the kinetic burning when applied to Zone II char combustion: the elim-
rate and the diffusion limitations are important, as is ination from the analysis of particles with measured
expected to be the case for the experiments reported burning rates at or above the calculated diffusion limit
here. Under these conditions, the diffusion rate of can significantly bias the result. Imagine a popula-
oxygen to the particle needs to be accounted for, as tion of particles burning, on average, very close to the
well as the mix of external surface reaction and inter- diffusion limit. Particle-to-particle variations in reac-
nal pore reaction. Note that this latter difficulty does tivity as well as experimental uncertainty will scatter
not, in fact, simply devolve into the classic Thiele ki- the measured burning rates around the diffusion limit.
netic analysis for reaction in homogeneous, porous Now, if all points at or above the diffusion limit are
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 717

eliminated from the analysis, the resulting population fidence intervals covering the full range of parameter
of points will, on average, be burning more slowly values (i.e., 0  A  ∞, 0  E  ∞, and 0  n  1).
than the original population. In other words, the aver- Several different forms of the Langmuir–Hinshel-
age particle reactivity derived from the analysis will wood two-step kinetic expression were considered in
be lower than the actual average reactivity of the orig- this study. The best performance was found for what
inal population. The fact that the determination of we refer to as an nth-order Langmuir–Hinshelwood
pO2 ,s is subject to significant uncertainty exacerbates equation
this problem. n
k2 k1 pO
2 ,s
To provide a robust, quantitative means of com- q= n , (7)
paring the fits of different char kinetics equations to k1 pO +
,s k2
2
the data measured in this study, a nonlinear regres- with the k’s, as before, expressing Arrhenius depen-
sion technique was developed for determining the ki- dence on the particle temperature. One may think
netics coefficients. The use of nonlinear techniques of this expression as a description of adsorption–
and genetic algorithms to determine Arrhenius kinetic desorption kinetics in which one does not specify the
constants has become increasingly common over the reaction order of the O2 adsorption step (as suggested
past decade [45–50], but to our knowledge has not in Ref. [52]). This form is also similar to, but not quite
been previously applied to char oxidation kinetics. In the same, as that which results from applying the con-
addition to avoiding the difficulties of linearized ap- ventional two-step Langmuir–Hinshelwood equation
proaches for determining char kinetics near a limiting to Zone II combustion [44] and approximates the ef-
condition, such as the diffusion limit, nonlinear re- fective kinetic behavior for Langmuir–Hinshelwood
gression techniques allow one to readily incorporate kinetics when applied to char particles with a mildly
a number of different kinetic expressions for fitting. heterogeneous reactivity distribution [41].
In addition, the nonlinear regression technique yields
quantitative confidence intervals that indicate the trust 3.4. Char burnout model
one can have in the parameter values that are derived
from the technique. To check the derived kinetic rates for consistency
To perform nonlinear regression, trial values of with the measured char burnout profiles, the equa-
A, E, and n are used to determine q and pO2 ,s , us- tions describing particle mass and temperature were
ing Eq. (2) as described above. Rather than solving integrated in the reactor to obtain temperature and
Eq. (1), however, we calculate a residual from Eq. (1) conversion profiles for each coal under intermediate-
for each particle. A modified Levenberg–Marquardt temperature conditions. The differential equation de-
method, based on the MINPACK routines LMDIF scribing the evolution of particle mass is
and LMDER, by Moré et al. [51], is used to opti-
mize the values of A, E, and n by minimizing the sum dmp π dp2 qWc
= ; (8)
of the squares of the residuals. With this approach, q dz vp
values are not actually determined as part of the opti- Eq. (1) describes the evolution of particle tempera-
mization procedure, so there is no possibility of bias ture.
from characteristic curves as in the linear regression The burnout model relies on numerous parame-
approach. Also, different expressions for the kinetic ters, many of which have significant uncertainty. To
rate are simply applied as different functions from determine how this parameter uncertainty affects the
which to calculate q, given trial values of the relevant predictions of the burnout model, the input parame-
kinetic constants. A and E are constrained to be pos- ters and the output profiles are modeled as stochastic
itive and n is limited to the range from zero to one variables. The input parameters are each modeled us-
using the transformations A = exp θ0 , E = exp θ1 , ing a normal distribution with a mean equal to the
and n = 1/(1 + exp θ2 ). Confidence intervals are cal- estimated value of the parameter and a standard de-
culated for each of the fitted kinetics constants using a viation equal to half the estimated uncertainty in the
linear approximation analogous to that used for linear variable. (Parameters with a lower bound, such as re-
regressions. action rates, are modeled using a lognormal distribu-
The technique described here has the advantage tion.) The uncertainties in the input parameters are all
that it correctly handles particles that are in the dif- assumed to be independent; thus, uncertainties in sev-
fusion limit, since the equations are valid throughout eral input variables are modeled simultaneously using
the range of conditions from the kinetic limit to the the same number of independent random variables.
diffusion limit. If a regression is performed using a The output variables in the model are also stochas-
dataset containing particles that are all burning at or tic variables. To study how uncertainty propagates
near the diffusion limit, the algorithm will return con- through the model, a number of simulations are run
718 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729

Table 2
Parameters used in the char burnout model, with estimated values and uncertainties
Symbol Parameter Estimate 2σ (%)
ρp,i Initial char density (kg/m3 ) 700.0 (hv) 11 (hv)
550.0 (eb) 5 (eb)
Tp,i Initial char temperature (K) 1900.0 10
dp,i Initial char diameter (µm) 115.0 10
Tg (z) Gas temperature (K) 1300–1685 4
CO2 /CO Heterogeneous product ratio 0.02pO0.21 e3070 K/Tp 100a
,s
2
pO2 Bulk gas oxygen content (mol%) 12.0 2
Tw Wall temperature (K) 500.0 40
ρash Ash particle density (kg/m3 ) 2250.0 18
Yc,i Initial char carbon content (mass%) 73.7 18
a Simulated using a lognormal distribution.

in which the input parameters are varied randomly.

The effects of these variations on the output variables
are analyzed. This technique is superior to traditional
sensitivity analysis since it captures nonlinear effects
and parameter interactions.
The stochastic variables in the model are para-
meterized using polynomial chaos expansions [53].
These expansions allow one to write a stochastic vari-
able as a series of modes, similar to a Fourier series
expansion. Terms in the polynomial chaos expansion
act to skew or shift the distribution of the variable.
The more terms that are included in the expansion,
the more accurately the expansion is expected to rep-
resent the true distribution of the variable. The poly-
nomial chaos expansion is used here because of its
favorable convergence properties.
Values of nine model input parameters are shown
in Table 2, along with estimated uncertainties. The un-
certainties are given as 95% confidence intervals, or Fig. 4. Time lapse photographs of Highvale coal combustion
twice the standard deviation, σ , of the distribution. in Sandia’s entrained flow reactor under the intermediate gas
The distributions are sampled using Sandia’s Latin temperature conditions. The oxygen content in the bulk gas
Hypercube sampling program [54]. Latin Hypercube is indicated below each photograph. The dashed line gives
is a stratified sampling technique that is more efficient a qualitative indication of where coal devolatilization has
ended and where char combustion is beginning. A length
than pure random sampling and provides faster con-
scale is indicated to the right of the photographs.
vergence.

temperature increases (as evidenced by the incandes-

4. Results and discussion cent intensity and color) as the oxygen concentration
increases, and char burnout (evidenced by the end-
4.1. Flow reactor photographs and observations point of the incandescent particle traces) occurs much
faster. It is also interesting to note that the bright lu-
Fig. 4 shows photographs of the Highvale coal minous emission from oxidizing soot that is evident
particle combustion history for the intermediate- in the volatiles flames (low in the furnace) at low
temperature condition of the entrained flow reactor. oxygen concentrations disappears at higher oxygen
The same trends with oxygen content evident in these levels.
photographs were also apparent for the eastern bi- The apparent enhancement of the devolatilization
tuminous coal. As the bulk oxygen concentration rate with increasing oxygen content probably results
increases, devolatilization occurs more rapidly and from two factors: the closer proximity of the volatiles
incandescence from the burning char particles is visi- flame to the coal particle and the higher tempera-
ble lower in the reactor. The char particle combustion ture of the volatiles flame. A closer proximity of
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 719

the volatiles flame at high oxygen levels is expected

on the basis of diffusional constraints and the ex-
istence of the flame sheet at a standoff distance at
which the emitted volatiles and the oxygen in the
surrounding bulk gas diffuse toward each other in
stoichiometric proportions [55]. Because soot incep-
tion is a relatively slow process in comparison to the
small-molecule gas-phase reactions in the volatiles
flame, the lack of observable soot thermal emission
in the oxygen-enriched environments probably indi-
cates that there is insufficient residence time for soot
to form in the fuel-rich region between the particle
and the surrounding volatiles flame. Adiabatic flame
temperature calculations, using methane as a surro-
gate for the complex mixture of volatile gases evolved
from the coal, show that the characteristic volatiles
flame temperature in the entrained flow reactor en- Fig. 5. Fractional coal mass remaining as a function of
vironment increases from 2190 K at 6 mol% O2 to residence time in the entrained flow reactor, for different
2450 K at 12 mol% O2 to 2860 K at 36 mol% O2 . bulk oxygen concentrations and coal types, at the intermedi-
ate-temperature burner condition. Exponential curve fits to
4.2. Coal burnout and elemental release rates the data points are shown.

Char collection was performed at the intermediate- height in the reactor, extrapolating the fit back to the
temperature condition for each oxygen concentration base of the furnace, and then integrating the inverse
at three different reactor heights for the Highvale coal of the particle velocity over the reactor height. The
and at five different heights for the eastern bitumi- use of higher oxygen concentrations clearly results in
nous coal. These samples were submitted to ultimate faster char burnout, as was also evident from direct
and proximate analysis at a commercial analytical visual observations and the time-lapse photographs
laboratory. The Highvale samples also underwent in- of the coal combustion process. A few of the char
ductively coupled plasma (ICP) analysis of the con- samples, collected low in the reactor, have significant
tent of the refractory metals Si, Al, and Ti. From volatile contents and were evidently collected prior
these analyses, the fractional burnout of the two coals to complete coal devolatilization. From the trends in
could be determined as a function of residence time the burnout data, the high-temperature dry, ash-free
in the reactor, and the fractional elemental release (DAF) volatiles loss is approximately 50 wt% for the
from the coal could be determined as a function Highvale coal and 40 wt% for the eastern bituminous
of burnout. To estimate char burnout for the High- coal, consistent with existing databases for coals of
vale coal, individual refractory metal and total ash this rank [56,57].
tracer techniques were examined. The different meth- Fig. 6 shows the fractional elemental release of
ods typically yielded values within a few percent of the two coals, as a function of coal burnout. Previ-
one another, with a small bias toward larger appar- ous research under conventional combustion condi-
ent burnout values when using titanium as the tracer. tions has shown that the elemental release is a unique
Because the burnout calculation assumes no loss of function of the total mass release when both are ex-
these metals and ash species through volatilization pressed on a DAF basis [25,58]. For Highvale coal,
(and thus is really a lower limit to the actual burnout), the trends of early hydrogen and oxygen release and
the titanium-based burnout is assumed to be superior delayed carbon, nitrogen, and sulfur release are in
to use of the other analyzed metal species or the total good agreement with the elemental release rates pre-
ash content, and is used here. For the eastern bitumi- viously determined for subbituminous coals in the
nous coal, only information on the total ash content same entrained flow reactor [25]. Similarly, the nearly
was available, so this was used as the tracer to deter- constant carbon release, faster hydrogen and oxygen
mine the mass burnout. release, and more uniform nitrogen and sulfur release
Fig. 5 shows the burnout profiles of both coals of the high-volatile eastern bituminous coal are con-
at the intermediate-temperature condition as a func- sistent with previous trends [25,58]. For both coals,
tion of the oxygen content of the flow. The residence there is no significant sensitivity of the DAF elemen-
time of the particles within the reactor has been cal- tal release trends to combustion in 6% versus 36% O2 ,
culated by applying a polynomial curve fit to the despite the substantial differences in particle temper-
measured mean particle velocities as a function of ature and burning rate over this range of conditions.
720 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729

Fig. 6. DAF elemental release versus the DAF mass loss for combustion of Highvale subbituminous and high-volatile eastern
bituminous coal in environments ranging from 6 to 36% O2 . The solid line indicates the unity relationship for elemental release
equal to mass release. The vertical, rectangular box indicates the range of mass release corresponding to complete volatile loss
for the two coals. The dashed lines represent simple polynomial curve fits to the experimental data.

The scattered data and slow apparent release of sul- high residence times for both coals under all con-
fur for Highvale coal probably reflect measurement ditions; the falloff with residence time is somewhat
difficulties associated with the large amount of sulfur larger for the eastern bituminous coal.
associated with its ash. For Highvale coal combustion in 6 and 12% O2
at the intermediate temperature condition, the actual
4.3. Measured temperature and size distributions distribution of char particle sizes as a function of
burnout in the flow reactor was measured by Mal-
As has been previously observed in a number of oney et al. [59] and shows the expected decrease in
studies, different char particles showed a significant mean char particle size (for near-diffusion-limit com-
variation in temperature at any given sampling height bustion) from 100 µm for young chars down to 85 µm
and burner condition, for any given coal. A portion of for 40% char burnout. In contrast, the average char
this variation may be attributed to actual differences particle diameters measured with the in situ optical
in particle temperatures (resulting from the heteroge- technique are larger than this and show an increase in
neous distribution of intrinsic reactivity, pore volume particle size at short residence times, followed by a
structure, and ash content within the particle popu- decrease in size at later times. The larger sizes mea-
lation) and a portion results from measurement un- sured in situ result from thermal expansion of the hot
certainty. In Fig. 7, average measured char particle char and the inherent bias of the laser-triggered op-
temperatures and diameters for the intermediate tem- tical pyrometry technique to larger particles [13]. It
perature flow condition are shown as a function of res- should be noted that the distribution of in situ mea-
idence time in the flow reactor. The average particle sured particle sizes is quite wide for either coal at any
temperatures for the Highvale coal are approximately given measurement location, with standard deviations
1800, 2000, 2200, and 2400 K at the 6, 12, 24, and varying from 20 to 35 µm. Therefore, the size distri-
36% oxygen conditions, respectively. For the east- butions at all heights and flow conditions show sub-
ern bituminous coal, particle temperatures are slightly stantial overlap, even for the Highvale coal burning in
lower: 1800, 1950, 2100, and 2300 K for the same 36% O2 , in which the measured mean diameters are
conditions. The average particle temperatures tend to larger than for the other oxygen levels. Calculations
increase at low residence times and then decrease at show that the derived surface-specific particle burn-
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 721

Fig. 7. Average measured char particle temperatures and sizes for the Highvale (a) and eastern bituminous (b) coals. The bars
show the range of plus/minus one standard deviation of the measured temperature distribution around the average temperature.

Fig. 8. Measured versus diffusion-limit burning rates for the Highvale (a) and eastern bituminous (b) coals. Points above the
solid line have measured rates greater than the calculated diffusion-limit rate. The ellipses show 95% confidence regions for
typical points.

ing rates are relatively insensitive to particle size in ing rate, because the diffusion-limit burning rate is
this size range (they are primarily dependent on parti- inversely proportional to the particle diameter (which,
cle temperature). as previously noted, has a high uncertainty).
The representation of the data shown in Fig. 8 is
4.4. Char burning rate trends useful for confirming the accuracy of the boundary
layer diffusion calculation and for evaluating the con-
Fig. 8 shows the experimentally measured burning ditions under which the determination of meaningful
rates, for all three reactor temperature profiles, plot- kinetic rate information from the data may be possi-
ted against the diffusion-limit burning rates for both ble. The good agreement that is evident between the
the Highvale and the eastern bituminous chars. The calculated diffusion-limit burning rate and the mea-
diffusion-limit burning rates are calculated by solving sured burning rate as the bulk oxygen concentration
Eqs. (1) and (2) for both q and Tp using the measured approaches zero (forcing the char to Zone III combus-
particle sizes, while setting pO2 ,s = 0. The 95% con- tion) provides direct evidence of the accuracy of the
fidence regions shown in these plots are typical for boundary layer diffusion calculation in this study. For
these data. The uncertainty in the diffusion-limit burn- combustion conditions in which the vast majority of
ing rate is much larger than that of the measured burn- data points straddle the diffusion limit, it is not possi-
722 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729

Fig. 9. Representative, median ks ’s and determination of the kinetic constants for the Highvale (a) and the eastern bituminous (b)
coals using the method of Hurt and Mitchell for nth-order Arrhenius kinetics with an assumed reaction order of 0.5. For the
Highvale coal, fits are shown for the entire set of median points and for those points with particle temperatures greater than
1830 K. The dotted lines show the 95% confidence prediction regions for the fits.

Table 3
Linear regression results for the nth-order Arrhenius rate equation
Fit A (mol/s m2 atmn ) E (kJ/mol) n Residualsa
Highvale coal
Free fit 1073 ± 48 73.0 ± 0.9 0.02 ± 0.01 47
Fixed n 1498 ± 120 53.4 ± 1.5 0.50 157
Hurt and Mitchell 1989 ± 639 57.7 ± 5.2 0.50 121
Hurt and Mitchell (Tp > 1830 K) 306 ± 121 23.2 ± 7.0 0.50 43
Eastern bituminous coal
Free fit 615 ± 22 63.8 ± 0.7 0.04 ± 0.01 38
Fixed n 238 ± 16 26.6 ± 1.1 0.50 132
Hurt and Mitchell 94 ± 26 10.4 ± 4.2 0.50 45
a Proportional to the sum of the squared fit residuals divided by degrees of freedom of the fit residuals.

ble to extract any kinetic information. Fig. 8 shows ered by making measurements at high oxygen lev-
that many of the particles are burning at or above els. The smooth overlap of the data from oxygen-
the calculated diffusion-limit rate for combustion at enriched atmospheres with that from reduced oxy-
the lower oxygen concentrations, where most previ- gen atmospheres suggests that there is no dramatic
ous char kinetic studies have been performed. This change in the burning behavior of the char particles
tendency is particularly evident for the Highvale coal. over the investigated range of conditions. However, as
These findings are consistent with earlier char kinetic noted previously, the applicability of the single-film
studies in this same flow reactor under similar tem- model to derive char burning rates becomes increas-
perature conditions with 6 and 12% oxygen, in which ingly uncertain as the oxygen content in the bulk gas
kinetic rate information could not be extracted from increases. Interestingly, the observed trend of burn-
the data for subbituminous coals [60]. The eastern bi- ing rate decreasing relative to the diffusion limit at
tuminous coal, which is less reactive, has a smaller, elevated oxygen levels is the opposite of what would
but still significant, percentage of char particles burn- be expected if, in fact, oxidation of carbon monoxide
ing in the diffusion limit at the 6 and 12% oxygen became significant in the char boundary layer as the
conditions. oxygen level increased [28,61].
It is also apparent from these plots that increasing
oxygen concentration in the bulk gas moves the char 4.5. Char kinetic rate
particles away from the diffusion limit, despite the
increased particle temperatures in oxygen-enriched 4.5.1. Linear fits of nth-order kinetics
atmospheres. This finding implies that the increase Fig. 9 and Table 3 show the results of a linear least-
in the kinetic burning rate in the oxygen-enriched squares regression of the char burning rate data for the
atmospheres fails to offset the increase in the dif- Highvale and eastern bituminous coals, using the Hurt
fusional flux of oxygen to the particle. This trend and Mitchell approach for fitting the nth-order Arrhe-
also suggests that better kinetics data can be gath- nius expression with an assumed reaction order of 0.5.
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 723

Fig. 10. Linear, least-squares fit of nth-order Arrhenius kinetics for the entire population of particle datapoints for the High-
vale (a) and the eastern bituminous (b) coals, with an assumed reaction order of 0.5.

Two different Hurt and Mitchell fits are indicated for in Table 3. The best-fit activation energies, in this
the Highvale coal, one that uses all of the median data case, are around 70 kJ/mol, which is consistent with
points and one that selectively excludes those median previous measurements of Zone II combustion kinet-
points corresponding to measurement conditions that ics [25,37]. A comparison of the relative fit residuals
had a preponderance of data points in the diffusion shows that the free fit is a much better representation
limit (which includes all of the 6% O2 data and about of the data than the forced fit with an assumed reac-
half of the 12% O2 data). For those conditions, the tion order of 0.5.
median datapoints of the non-diffusion-limit particles
are weakly reacting (probably corresponding to dense 4.5.2. Nonlinear fits of nth-order kinetics
carbon or ash-dominated particles) and inappropri- Table 4 shows the results of nonlinear fits of vari-
ately bias the fit downward. If these low medians are ous kinetic expressions to the experimental data. The
removed from consideration (as has been the practice free fit of the nth-order Arrhenius expression to all
when applying this type of fit), the resultant High- of the data yields best-fit activation energies and re-
vale coal fit shows a similar activation energy to that action orders for both coals that are nominally equal,
determined for the eastern bituminous coal and ap- 45 kJ/mol and 0.18, respectively. By refitting the data
propriately shows that the Highvale coal has a higher with a common activation energy of 45.0 kJ/mol and
reactivity over the entire range of investigated temper- a common reaction order of 0.175, the preexponen-
atures. For both coals, the apparent activation energies tials may be compared to elucidate the relative reac-
determined by this linear fit are very low (23 kJ/mol tivity of the coals. This analysis shows that, on av-
for Highvale and 10 kJ/mol for the eastern bitumi- erage, the Highvale coal is 40% more reactive than
nous coal). As will be shown later, the low activation the eastern bituminous coal. This finding is consistent
energy results, in part, from forcing the kinetic fit to with previous measurements of the char kinetic rates
have a reaction order of 0.5. of subbituminous coals relative to Pittsburgh seam
A linear fit of the nth-order Arrhenius expres- high-volatile bituminous coal [25,37].
sion with an assumed reaction order of 0.5 to all the Fig. 11 shows the results of a series of nonlin-
non-diffusion-limit particle kinetic data is shown in ear fits to all of the experimental data of the nth-
Fig. 10. The trending of the characteristic curves is order Arrhenius expression for fixed values of n. The
clearly evident. Also evident is the dearth of non- plot of residuals shows a distinct minimum for n =
diffusion-limit points at low particle temperatures 0.17–0.18. Also noteworthy is the dramatic decrease
(corresponding to combustion in 6 and 12% O2 ), par- in the activation energy with increasing oxygen ex-
ticularly for the Highvale coal. As shown in Table 3, ponent, reaching zero for the Highvale coal for n ∼ 1.
the kinetic constants determined by the linear fit of The trends evident in Fig. 11 explain the observed dif-
the individual particle data points are quite similar to ference in the kinetic fit constants for linear versus
those determined by the Hurt and Mitchell methodol- nonlinear fitting of the char burning rate data. With
ogy. Indeed, the predicted ks values are within 15% the exclusion of diffusion-limit data points in the lin-
of one another over the particle temperature range of ear fitting procedure, the fit becomes locally biased
1800–2500 K (encompassing almost all of the data). toward the unreactive portion of the particle popula-
If one performs a linear, free fit of the nth-order tion (as previously discussed), having a net effect of
Arrhenius expression to all of the non-diffusion-limit forcing the fit to a higher activation energy. As shown
data, the best-fit value for n is near zero, as shown in Fig. 11, the best fit for a constraint of high activa-
724 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729

Table 4
Nonlinear regression results for the nth-order Langmuir–Hinshelwood (LH) and nth-order Arrhenius rate equations
Fit A1 (mol/s m2 atmn ) E1 (kJ/mol) n A2 (mol/s m2 ) E2 (kJ/mol) Residualsa
Highvale coal
Arrhenius 475 ± 30 45.9 ± 1.3 0.17 ± 0.01 421
Arrhenius, 24 + 36% O2 5230 ± 639 79.1 ± 2.1 0.46 ± 0.02 329
Arrhenius, 6 + 12% O2 4933 ± 939 83.2 ± 3.2 0.10 ± 0.03 258
LH 93 ± 26 0.1 ± –b 0.30 ± 0.04 26.2 ± 15.2 109.9 ± 7.5 400
LH, n = 1 2501 ± 1683 0.1 ± –b 1.00 1.1 ± 0.1 66.8 ± 2.3 453
LH-Essenhighc 7617 ± 392 31.4 1.00 7.7 ± 0.1 100.0 566
Eastern bituminous coal
Arrhenius 344 ± 16 45.5 ± 1.0 0.18 ± 0.01 265
Arrhenius, 24 + 36% O2 1086 ± 84 54.4 ± 1.3 0.50 ± 0.02 228
Arrhenius, 6 + 12% O2 1401 ± 227 69.1 ± 2.7 0.10 ± 0.02 168
LH 61 ± 17 0.5 ± –b 0.32 ± 0.04 20.0 ± 11.8 107.4 ± 7.3 260
LH, n = 1 1192 ± 742 0.0 ± –b 1.00 0.6 ± 0.1 60.7 ± 2.0 295
LH-Essenhighc 2618 ± 97 31.4 1.00 8.1 ± 0.2 100.0 413
a Proportional to the sum of the squared fit residuals divided by degrees of freedom of the fit residuals.
b For these fits, the confidence interval for E was 0–∞.
1
c Activation energies taken from Ref. [39], E = 31.4 kJ/mol and E = 100.0 kJ/mol.
1 2

found for the fit to the oxygen-enriched data is more

compatible with the classic Thiele analysis, as previ-
ously discussed.
Table 4 also shows that nonlinear fits of the nth-
order Arrhenius expression to the reduced oxygen
data (6 and 12% O2 ) yield higher activation ener-
gies (83 and 69 kJ/mol) than are found for fits to
the complete dataset. As mentioned previously, acti-
vation energies of 70–80 kJ/mol are consistent with
past measurements of coal char kinetics in reduced
oxygen atmospheres [25,37]. The best-fit reaction or-
der for the reduced oxygen data is 0.1 for both coals,
which explains why the fit to all of the data yields a re-
action order that is significantly lower than the value
of 0.5 determined for the oxygen-enriched combus-
tion conditions.
Fig. 11. Best-fit A and E and relative residuals for nonlinear
fits of the nth-order Arrhenius model versus choice of n. 4.5.3. Nonlinear fits of Langmuir–Hinshelwood
kinetics
tion energy is a reaction order of zero, as evidenced in Fig. 12 shows the results of a nonlinear free fit of
the linear fit results. Similarly, constraining the reac- the nth-order Langmuir–Hinshelwood expression to
tion order to be 0.5 forces the activation energy to be all of the experimental data. Because the char burning
relatively low, as also seen in the linear fit results. rates (q’s) that are being fit depend on both particle
As shown in Table 4, nonlinear fits of the nth-order temperature and surface oxygen partial pressure, the
Arrhenius expression to only the oxygen-enriched results of the fit cannot be conveniently shown on an
data yield higher activation energies than are found Arrhenius plot of q vs 1/Tp . Instead, we resort to giv-
for fits to the complete dataset (79 and 54 kJ/mol, ing a comparison of the predicted versus the measured
vs 46 kJ/mol). In addition, the apparent reaction or- burning rate. The best-fit kinetic parameters, shown
ders are considerably higher (0.5 vs 0.2). Because in Table 4, yield an activation energy for k1 (the step
the oxygen-enriched combustion data do not contain representing adsorption) of essentially zero and for k2
many datapoints at or above the diffusion limit, an ar- (representing desorption) of ∼110 kJ/mol. The best-
gument can be made to weight kinetic fits of these fit reaction order for the adsorption step is 0.3, similar
data over fits to the complete dataset, even when ap- to the reaction order of 0.2 found for the nonlinear
plying this to char combustion in conventional envi- fit of the nth-order Arrhenius expression to all of the
ronments. In addition, the reaction order of 0.5 that is data. If the reaction order of the adsorption step is set
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 725

Fig. 12. Predicted versus measured burning rate for a nonlinear free fit of the nth-order Langmuir–Hinshelwood expression for
Highvale (a) and eastern bituminous (b) coal chars.

imental data in the transition region (i.e., “roll-off”)

between desorption control (at lower temperatures)
and adsorption control (at higher temperatures). Thus,
there is considerable curvature in the dependence of
log q on 1/Tp in the temperature range of interest.
The nth-order Arrhenius expression can provide only
a linear dependence of log q on 1/Tp . The improved
fit of the Langmuir–Hinshelwood relation indicates
that the experimental data demonstrate some of this
curvature, which may be important when extrapo-
lating the fits to lower or higher temperatures than
Fig. 13. Comparison of nth-order Langmuir–Hinshelwood investigated here.
equation to the nth-order Arrhenius expression for two dif-
ferent surface oxygen partial pressures, using best-fit coef- 4.6. Results of burnout calculations
ficients for the eastern bituminous coal. The adsorption and
desorption limits of the Langmuir–Hinshelwood expression Results of the burnout model calculations when
are indicated, as are the nth-order Arrhenius rates and the using the best-fit nth-order Langmuir–Hinshelwood
range of experimentally measured particle temperatures. equation are shown in Fig. 14. These calculations
were initiated at a residence time of 25 ms, when
to one, as in the traditional Langmuir–Hinshelwood devolatilization is complete. The predictions gener-
equation, the activation energy for k1 is again forced ally show a good correlation with the measured dry,
to zero, the activation energy for k2 is reduced to ash-free mass loss rate under the different combustion
60–70 kJ/mol, and the residuals show that the fit is in- conditions, although the burnout rate of the Highvale
ferior to the fit when n is allowed to vary. If one goes coal under oxygen-enriched conditions is overesti-
further and adopts the activation energies proposed by mated. Note that this model does not include mech-
Essenhigh and Mescher [39] for the adsorption and anisms for ash inhibition, statistical kinetics, or ther-
desorption steps, fitting just the preexponentials for mal annealing of the char [62], so the burnout rate at
the two kinetic rates, the quality of the fit deteriorates high extents of conversion tends to be overestimated.
significantly. Char particle temperatures are predicted well under
Table 4 shows that the normalized residuals from oxygen-enriched combustion conditions, but tend to
the fit of the nth-order Langmuir–Hinshelwood equa- be underpredicted, particularly at long times, for re-
tion are slightly lower than those from the nth- duced oxygen combustion conditions. This underpre-
order Arrhenius fit to all of the data, indicating the diction is caused, in part, by the inherent bias of the
Langmuir–Hinshelwood expression yields a superior optical diagnostic measurement system to capturing
fit. Fig. 13 gives a comparison of the best-fit nth- data on larger particles (i.e., with this bias, the mea-
order Arrhenius expression and nth-order Langmuir– sured particle temperatures at long residence time cor-
Hinshelwood equation for the eastern bituminous coal respond to larger initial particles than the 115 µm as-
char, for two characteristic values of surface oxy- sumed in the burnout calculation).
gen partial pressures. As is evident in the figure, Samples of the calculated uncertainty in the burn-
the Langmuir–Hinshelwood fit has placed the exper- out profiles are shown in Fig. 15. To calculate the
726 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729

Fig. 14. Results of burnout model simulations for the Highvale (a) and the eastern bituminous (b) coals, using the best-fit kinetic
parameters for the nth-order Langmuir–Hinshelwood equation.

Fig. 15. Burnout model simulations for the Highvale (a) and the eastern bituminous (b) coals for combustion in 12% O2 . The
nominal burnout model result is shown as a solid line, and dashed lines indicate the 95% confidence limits for the simulation,
based on the parameter uncertainties given in Table 2.

uncertainty, all nine parameters shown in Table 2 still actively burning. This effect is particularly dra-
were varied simultaneously over 24,000 simulations. matic for the Highvale coal, which is calculated to
Several runs were done with different random num- undergo complete combustion in the 120-ms time
ber seeds to check for convergence. The devolatiliza- window.
tion mass loss is assumed to be known for these Fig. 16 shows how uncertainties in the different
simulations, so the uncertainty in char mass burnout burnout model parameters contribute to the uncer-
necessarily increases with time, demonstrating a cu- tainty in the calculated profiles of char particle tem-
mulative effect. For char particle temperature, the perature and burnout. Uncertainties in the initial char
uncertainty initially decreases during the peak com- density and initial particle size are the major con-
bustion period and then expands to large values as tributors to the Highvale coal profiles and to the ini-
some simulated particles burn out while others are tial portion of the burnout of the eastern bituminous
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 727

Fig. 16. Stacked plots of the contributing factors to the burnout model variance for the Highvale (a) and the eastern bituminous (b)
coals for combustion in 12% O2 .

coal. Note that in this study the relative uncertainty crease the char combustion temperature and to re-
in the initial char density of Highvale is over twice duce the char burnout time, as expected. The opti-
that of the eastern bituminous coal, which accounts cal kinetic data, interpreted with a single-film ox-
for at least some of the difference in the impact of idation model, demonstrate increasing kinetic con-
this factor on the simulation results. With the lower trol in enriched-oxygen combustion, despite the faster
reactivity of the eastern bituminous coal, uncertainty particle combustion rates. Both linear regression and
in the gas temperature contributes significantly to the nonlinear regression have been used to fit the op-
char temperature profile and to the latter portion of tical kinetic data over the entire investigated range
char burnout. Uncertainty in the bulk oxygen content of oxygen concentrations, as well as subsets of that
also significantly influences the latter portion of char range. Fits with both a simple nth-order Arrhenius
burnout of the eastern bituminous coal. Uncertainty in expression and an nth-order Langmuir–Hinshelwood
the initial particle temperature only contributes signif- expression were equally successful, yielding apparent
icantly to the char temperature profiles, and its effect reaction orders of 0.2 and 0.3, respectively, when ap-
rapidly becomes negligible as the particles heat up plied over all of the data. Local fits of the nth-order
and begin to burn. Arrhenius expression to the conventional and oxygen-
The burnout simulation results presented in Figs. 15 enriched atmospheres gave apparent reaction orders
and 16 highlight the need to control and accurately of 0.1 and 0.5, respectively. Char burnout rates and
measure all relevant coal, char, and combustion en- char particle temperatures are predicted reasonably
vironment parameters when making comparisons be- well when applying a char burnout model with the
tween burnout simulations and experimentally mea- derived kinetics.
sured particle temperatures and char burnout. In par-
ticular, a premium is placed on accurately knowing
the apparent density of the char and the initial char Acknowledgments
particle size, as was recently emphasized in a related
study [59]. This work was supported by the U.S. Depart-
ment of Energy through the National Energy Tech-
nology Laboratory’s Power Systems Advanced Re-
5. Conclusions search Program, managed by Dr. Robert Romanosky.
Mark Douglas, Eddy Chui, and Yewen Tan of CAN-
The combustion reactivities of Highvale subbitu- MET provided the size-classified Highvale and East-
minous coal char and an eastern United States high- ern Bituminous coal samples and arranged for chem-
volatile bituminous coal char were investigated in ical analysis of the collected Highvale char samples.
conventional and oxygen-enriched atmospheres us- Sandians Jimmy Ross and Gilbert Hofacker assisted
ing Sandia’s optical entrained flow reactor. Oxygen- in the operation of the entrained flow reactor and in
enriched combustion was found to significantly in- char collection. Sandia is a multiprogram laboratory
728 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729

operated by Sandia Corporation, a Lockheed Martin [26] J.J. Murphy, C.R. Shaddix, Effects of Stefan flow on
Company, for the United States Department of En- heat transfer from reacting carbon particles, Report No.
ergy under contract DE-AC04-94-AL85000. SAND2002-8072, Sandia National Laboratories, De-
cember 2003.
[27] R.E. Mitchell, R.J. Kee, P. Glarborg, M.E. Coltrin,
Proc. Combust. Inst. 23 (1990) 1169–1176.
References
[28] A. Makino, Combust. Flame 90 (1992) 143–154.
[29] L. Tognotti, J.P. Longwell, A.F. Sarofim, Proc. Com-
[1] R. Payne, S.L. Chen, A.M. Wolsky, W.F. Richter, Com-
bust. Inst. 23 (1990) 1207–1213.
bust. Sci. Technol. 67 (1989) 1–16.
[30] R.H. Hurt, J.M. Calo, Combust. Flame 125 (2001)
[2] D.G. Roberts, D.J. Harris, Energy Fuels 14 (2000) 483–
1138–1149.
489.
[31] J.R. Arthur, Trans. Faraday Soc. 7 (1951) 164–178.
[3] D.G. Roberts, D.J. Harris, T.F. Wall, Fuel 79 (2000)
[32] J. Bzowski, J. Kestin, E.A. Mason, F.J. Uribe, J. Phys.
1997–1998.
Chem. Ref. Data 19 (5) (1990) 1179–1232.
[4] W.C. Hecker, P.M. Madsen, M.R. Sherman, J.W. Allen,
[33] P.H. Paul, J. Warnatz, Proc. Combust. Inst. 27 (1998)
R.J. Sawaya, T.H. Fletcher, Energy Fuels 17 (2003)
495–504.
427–432.
[34] R.P. Brent, Comput. J. 14 (4) (1971) 422–425.
[5] L.D. Timothy, A.F. Sarofim, J.M. Béer, Proc. Combust.
[35] M.A. Field, D.W. Gill, B.B. Morgan, P.G.W. Hawksley,
Inst. 19 (1982) 1123–1130.
Combustion of Pulverised Coal, The British Coal Util-
[6] S. Niksa, R.E. Mitchell, K.R. Hencken, D.A. Tichenor,
isation Research Association, Leatherhead, England,
Combust. Flame 60 (1984) 183–193.
1967.
[7] B.J. Waters, R.E. Mitchell, R.G. Squires, N.M. Lauren-
deau, Proc. Combust. Inst. 22 (1988) 17–27. [36] M.F.R. Mulcahy, I.W. Smith, Rev. Pure Appl. Chem. 19
[8] M. Saito, M. Sadakata, M. Sato, T. Soutome, H. Mu- (1969) 81–108.
rata, Y. Ohno, Combust. Flame 87 (1991) 1–12. [37] R.H. Hurt, R.E. Mitchell, Proc. Combust. Inst. 24
[9] B.R. Stanmore, Y.-C. Choi, R. Gadiou, O. Charon, P. (1992) 1233–1241.
Gilot, Combust. Sci. Technol. 159 (2000) 237–253. [38] R.H. Essenhigh, Energy Fuels 5 (1991) 41–46.
[10] C.R. Monson, G.J. Germane, A.U. Blackham, L.D. [39] R.H. Essenhigh, A.M. Mescher, Proc. Combust.
Smoot, Combust. Flame 100 (1995) 669–683. Inst. 26 (1996) 3085–3094.
[11] J.J. Saastamoinen, M.J. Aho, J.P. Hämäläinen, R. Hern- [40] E.M. Suuberg, M. Wojtowicz, J.M. Calo, Proc. Com-
berg, T. Joutsenoja, Energy Fuels 10 (1996) 121–133. bust. Inst. 22 (1988) 79–87.
[12] T. Reichelt, T. Joutsenoja, H. Spliethoff, K.R.G. Hein, [41] R.H. Hurt, B.S. Haynes, Proc. Combust. Inst. 30 (2005)
R. Hernberg, Proc. Combust. Inst. 27 (1998) 2925– 2161–2168.
2932. [42] E.W. Thiele, Ind. Eng. Chem. 31 (1939) 916–920.
[13] T. Joutsenoja, J. Saastamoinen, M. Aho, R. Hernberg, [43] R.E. Mitchell, Proc. Combust. Inst. 28 (2000) 2261–
Energy Fuels 13 (1999) 130–145. 2270.
[14] R.J. Tidona, Combust. Flame 38 (1980) 335–337. [44] D. Fortsch, R.H. Essenhigh, U. Schnell, K.R.G. Hein,
[15] T.F. Wall, G.-S. Liu, H.-W. Wu, D.G. Roberts, K.E. Energy Fuels 17 (2003) 901–906.
Benfell, S. Gupta, J.A. Lucas, D.J. Harris, Prog. En- [45] N.H. Chen, R. Aris, AIChE J. 38 (4) (1992) 626–628.
ergy Combust. Sci. 28 (2002) 405–433. [46] V. Cerda, A. Cladera, J.M. Estela, Quím. Anal. 15
[16] D.G. Roberts, D.J. Harris, T.F. Wall, Energy Fuels 17 (1996) 341–350.
(2003) 887–895. [47] N. Brauner, M. Shacham, Chem. Eng. Process. 36
[17] D.A. Tichenor, R.E. Mitchell, K.R. Hencken, S. Niksa, (1997) 243–249.
Proc. Combust. Inst. 20 (1984) 1213–1221. [48] T.Y. Park, G.F. Froment, Comput. Chem. Eng. 22
[18] R.E. Mitchell, Combust. Sci. Technol. 53 (1987) 165– (1998) S103–S110.
186. [49] S.P. Asprey, Y. Naka, J. Chem. Eng. Jpn. 32 (1999)
[19] C.R. Shaddix, Proc. 33rd National Heat Transfer Con- 328–337.
ference, HTD99-282, ASME, New York, 1999. [50] L. Elliott, D.B. Ingham, A.G. Kyne, N.S. Mera, M.
[20] D.C. Collis, M.J. Williams, J. Fluid Mech. 6 (1959) Pourkashanian, C.W. Wilson, Combust. Sci. Tech-
357–384. nol. 175 (2003) 619–648.
[21] P.H. Paul, J. Warnatz, Proc. Combust. Inst. 27 (1998) [51] J.J. Moré, B.S. Garbow, K.E. Hillstrom, User guide for
495–504. MINPACK-1, Report No. ANL-80-74, Argonne Na-
[22] E. Croiset, K. Thambimuthu, A. Palmer, Can. J. Chem. tional Laboratories, 1980.
Eng. 78 (2000) 402–407. [52] E. Croiset, C. Mallet, J.-P. Rouan, J.-R. Richard, Proc.
[23] E. Croiset, K.V. Thambimuthu, Fuel 80 (2001) 2117– Combust. Inst. 26 (1996) 3095–3102.
2121. [53] R.G. Ghanem, P.D. Spanos, Stochastic Finite Ele-
[24] E.H. Chui, M.A. Douglas, Y. Tan, Fuel 82 (2003) 1201– ments: A Spectral Approach, Springer-Verlag, New
1210. York, 1991.
[25] R.E. Mitchell, R.H. Hurt, L.L. Baxter, D.R. Hardesty, [54] G.D. Wyss, K.H. Jorgensen, A user’s guide to LHS:
Compilation of Sandia coal char combustion data and Sandia’s Latin Hypercube Sampling software, Re-
kinetic analyses, Report No. SAND92-8208, Sandia port No. SAND98-0210, Sandia National Laboratories,
National Laboratories, 1992. 1998.
 J.J. Murphy, C.R. Shaddix / Combustion and Flame 144 (2006) 710–729 729

[55] I. Glassman, Combustion, third ed., Academic Press, [59] D.J. Maloney, E.R. Monazam, K.H. Casleton, C.R.
San Diego, CA, 1996. Shaddix, Proc. Combust. Inst. 30 (2005) 2197–2204.
[56] S. Niksa, Energy Fuels 5 (1991) 673–683. [60] R.H. Hurt, R.E. Mitchell, Proc. Combust. Inst. 24
[57] T.H. Fletcher, D.R. Hardesty, Compilation of Sandia (1992) 1243–1250.
coal devolatilization data, Report No. SAND92-8209, [61] A. Makino, C.K. Law, Proc. Combust. Inst. 21 (1986)
Sandia National Laboratories, 1992. 183–191.
[58] L.L. Baxter, R.E. Mitchell, T.H. Fletcher, R.H. Hurt, [62] R. Hurt, J.-K. Sun, M. Lunden, Combust. Flame 113
Energy Fuels 10 (1996) 188–196. (1998) 181–197.